Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

The matching effect of similar images is too different #102

Open
NPUXJH opened this issue Jan 10, 2024 · 5 comments
Open

The matching effect of similar images is too different #102

NPUXJH opened this issue Jan 10, 2024 · 5 comments

Comments

@NPUXJH
Copy link

NPUXJH commented Jan 10, 2024

Hello! I tried using LightGlue and encountered this problem: the matching effect of two similar scene images A and B and another image C is very different. For example, the matching effect of image A and image C is very good, but the matching effect of image B is very different. The match with image C is poor. Among them, the angle of view and scale of image A and image B are very similar. What is the reason for this?

@Phil26AT
Copy link
Collaborator

Hello @NPUXJH, i cannot answer this question without more information. Could you upload the images A, B, C here or send them to me privately?

@NPUXJH
Copy link
Author

NPUXJH commented Jan 11, 2024

Hello @NPUXJH, i cannot answer this question without more information. Could you upload the images A, B, C here or send them to me privately?

OK, I have sent the relevant pictures and results to the email address of your homepage, and I look forward to further replies. Thank you!

@Phil26AT
Copy link
Collaborator

Hi @NPUXJH

Indeed this is a weird failure case, I tested your example with different numbers of keypoints, and it seems the confidences get confused here, but only for 2K keypoints:

1K keypoints:
B_C_1K
2K keypoints:
B_C_2K
4K keypoints:
B_C_4K

Disabling early stopping results in:
B_C_2K_full

Going forward, I suggest to increase the early stopping confidence.

@noahzn
Copy link

noahzn commented Feb 26, 2024

@Phil26AT I have met the same problem. Increasing the early stopping confidence doesn't help.

@erdemkkok
Copy link

erdemkkok commented May 6, 2024

@Phil26AT hi,
have you trained your data before this matching results ?
if you trained your data before this results how can i train LightGlue weigths according to my data ? and
How can i fine-tune that ?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

4 participants